This adds handling of MSG_ZEROCOPY flag on transmission path:
1) If this flag is set and zerocopy transmission is possible (enabled
in socket options and transport allows zerocopy), then non-linear
skb will be created and filled with the pages of user's buffer.
Pages of user's buffer are locked in memory by 'get_user_pages()'.
2) Replaces way of skb owning: instead of 'skb_set_owner_sk_safe()' it
calls 'skb_set_owner_w()'. Reason of this change is that
'__zerocopy_sg_from_iter()' increments 'sk_wmem_alloc' of socket, so
to decrease this field correctly, proper skb destructor is needed:
'sock_wfree()'. This destructor is set by 'skb_set_owner_w()'.
3) Adds new callback to 'struct virtio_transport': 'can_msgzerocopy'.
If this callback is set, then transport needs extra check to be able
to send provided number of buffers in zerocopy mode. Currently, the
only transport that needs this callback set is virtio, because this
transport adds new buffers to the virtio queue and we need to check,
that number of these buffers is less than size of the queue (it is
required by virtio spec). vhost and loopback transports don't need
this check.
Signed-off-by: Arseniy Krasnov <avkrasnov@salutedevices.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
For tap device new skb is created and data from the current skb is
copied to it. This adds copying data from non-linear skb to new
the skb.
Signed-off-by: Arseniy Krasnov <avkrasnov@salutedevices.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This is preparation patch for MSG_ZEROCOPY support. It adds handling of
non-linear skbs by replacing direct calls of 'memcpy_to_msg()' with
'skb_copy_datagram_iter()'. Main advantage of the second one is that it
can handle paged part of the skb by using 'kmap()' on each page, but if
there are no pages in the skb, it behaves like simple copying to iov
iterator. This patch also adds new field to the control block of skb -
this value shows current offset in the skb to read next portion of data
(it doesn't matter linear it or not). Idea behind this field is that
'skb_copy_datagram_iter()' handles both types of skb internally - it
just needs an offset from which to copy data from the given skb. This
offset is incremented on each read from skb. This approach allows to
simplify handling of both linear and non-linear skbs, because for
linear skb we need to call 'skb_pull()' after reading data from it,
while in non-linear case we need to update 'data_len'.
Signed-off-by: Arseniy Krasnov <avkrasnov@salutedevices.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This adds support of MSG_PEEK flag for SOCK_SEQPACKET type of socket.
Difference with SOCK_STREAM is that this callback returns either length
of the message or error.
Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This reworks current implementation of MSG_PEEK logic:
1) Replaces 'skb_queue_walk_safe()' with 'skb_queue_walk()'. There is
no need in the first one, as there are no removes of skb in loop.
2) Removes nested while loop - MSG_PEEK logic could be implemented
without it: just iterate over skbs without removing it and copy
data from each until destination buffer is not full.
Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
Reviewed-by: Bobby Eshleman <bobby.eshleman@bytedance.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The read_skb hook calls consume_skb() now, but this means that if the
recv_actor program wants to use the skb it needs to inc the ref cnt
so that the consume_skb() doesn't kfree the sk_buff.
This is problematic because in some error cases under memory pressure
we may need to linearize the sk_buff from sk_psock_skb_ingress_enqueue().
Then we get this,
skb_linearize()
__pskb_pull_tail()
pskb_expand_head()
BUG_ON(skb_shared(skb))
Because we incremented users refcnt from sk_psock_verdict_recv() we
hit the bug on with refcnt > 1 and trip it.
To fix lets simply pass ownership of the sk_buff through the skb_read
call. Then we can drop the consume from read_skb handlers and assume
the verdict recv does any required kfree.
Bug found while testing in our CI which runs in VMs that hit memory
constraints rather regularly. William tested TCP read_skb handlers.
[ 106.536188] ------------[ cut here ]------------
[ 106.536197] kernel BUG at net/core/skbuff.c:1693!
[ 106.536479] invalid opcode: 0000 [#1] PREEMPT SMP PTI
[ 106.536726] CPU: 3 PID: 1495 Comm: curl Not tainted 5.19.0-rc5 #1
[ 106.537023] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS ArchLinux 1.16.0-1 04/01/2014
[ 106.537467] RIP: 0010:pskb_expand_head+0x269/0x330
[ 106.538585] RSP: 0018:ffffc90000138b68 EFLAGS: 00010202
[ 106.538839] RAX: 000000000000003f RBX: ffff8881048940e8 RCX: 0000000000000a20
[ 106.539186] RDX: 0000000000000002 RSI: 0000000000000000 RDI: ffff8881048940e8
[ 106.539529] RBP: ffffc90000138be8 R08: 00000000e161fd1a R09: 0000000000000000
[ 106.539877] R10: 0000000000000018 R11: 0000000000000000 R12: ffff8881048940e8
[ 106.540222] R13: 0000000000000003 R14: 0000000000000000 R15: ffff8881048940e8
[ 106.540568] FS: 00007f277dde9f00(0000) GS:ffff88813bd80000(0000) knlGS:0000000000000000
[ 106.540954] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 106.541227] CR2: 00007f277eeede64 CR3: 000000000ad3e000 CR4: 00000000000006e0
[ 106.541569] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 106.541915] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 106.542255] Call Trace:
[ 106.542383] <IRQ>
[ 106.542487] __pskb_pull_tail+0x4b/0x3e0
[ 106.542681] skb_ensure_writable+0x85/0xa0
[ 106.542882] sk_skb_pull_data+0x18/0x20
[ 106.543084] bpf_prog_b517a65a242018b0_bpf_skskb_http_verdict+0x3a9/0x4aa9
[ 106.543536] ? migrate_disable+0x66/0x80
[ 106.543871] sk_psock_verdict_recv+0xe2/0x310
[ 106.544258] ? sk_psock_write_space+0x1f0/0x1f0
[ 106.544561] tcp_read_skb+0x7b/0x120
[ 106.544740] tcp_data_queue+0x904/0xee0
[ 106.544931] tcp_rcv_established+0x212/0x7c0
[ 106.545142] tcp_v4_do_rcv+0x174/0x2a0
[ 106.545326] tcp_v4_rcv+0xe70/0xf60
[ 106.545500] ip_protocol_deliver_rcu+0x48/0x290
[ 106.545744] ip_local_deliver_finish+0xa7/0x150
Fixes: 04919bed94 ("tcp: Introduce tcp_read_skb()")
Reported-by: William Findlay <will@isovalent.com>
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Tested-by: William Findlay <will@isovalent.com>
Reviewed-by: Jakub Sitnicki <jakub@cloudflare.com>
Link: https://lore.kernel.org/bpf/20230523025618.113937-2-john.fastabend@gmail.com
This patch sets the skb owner in the recv and send path for virtio.
For the send path, this solves the leak caused when
virtio_transport_purge_skbs() finds skb->sk is always NULL and therefore
never matches it with the current socket. Setting the owner upon
allocation fixes this.
For the recv path, this ensures correctness of accounting and also
correct transfer of ownership in vsock_loopback (when skbs are sent from
one socket and received by another).
Fixes: 71dc9ec9ac ("virtio/vsock: replace virtio_vsock_pkt with sk_buff")
Signed-off-by: Bobby Eshleman <bobby.eshleman@bytedance.com>
Reported-by: Cong Wang <xiyou.wangcong@gmail.com>
Link: https://lore.kernel.org/all/ZCCbATwov4U+GBUv@pop-os.localdomain/
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
drivers/net/ethernet/mediatek/mtk_ppe.c
3fbe4d8c0e ("net: ethernet: mtk_eth_soc: ppe: add support for flow accounting")
924531326e ("net: ethernet: mtk_eth_soc: add missing ppe cache flush when deleting a flow")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This adds WARN_ONCE() and return from stream dequeue callback when
socket's queue is empty, but 'rx_bytes' still non-zero. This allows
the detection of potential bugs due to packet merging (see previous
patch).
Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This fixes appending newly arrived skbuff to the last skbuff of the
socket's queue. Problem fires when we are trying to append data to skbuff
which was already processed in dequeue callback at least once. Dequeue
callback calls function 'skb_pull()' which changes 'skb->len'. In current
implementation 'skb->len' is used to update length in header of the last
skbuff after new data was copied to it. This is bug, because value in
header is used to calculate 'rx_bytes'/'fwd_cnt' and thus must be not
be changed during skbuff's lifetime.
Bug starts to fire since:
commit 0777061657
("virtio/vsock: don't use skbuff state to account credit")
It presents before, but didn't triggered due to a little bit buggy
implementation of credit calculation logic. So use Fixes tag for it.
Fixes: 0777061657 ("virtio/vsock: don't use skbuff state to account credit")
Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This patch adds sockmap support for vsock sockets. It is intended to be
usable by all transports, but only the virtio and loopback transports
are implemented.
SOCK_STREAM, SOCK_DGRAM, and SOCK_SEQPACKET are all supported.
Signed-off-by: Bobby Eshleman <bobby.eshleman@bytedance.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Both of these functions have no effect when input argument is 0, so to
avoid useless spinlock access, check argument before it.
Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This adds small optimization for tx path: instead of allocating single
skbuff on every call to transport, allocate multiple skbuff's until
credit space allows, thus trying to send as much as possible data without
return to af_vsock.c.
Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Pointer to transport could be checked before allocation of skbuff, thus
there is no need to free skbuff when this pointer is NULL.
Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
Reviewed-by: Bobby Eshleman <bobby.eshleman@bytedance.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Link: https://lore.kernel.org/r/08d61bef-0c11-c7f9-9266-cb2109070314@sberdevices.ru
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This returns behaviour of SOCK_STREAM read as before skbuff usage. When
copying to user fails current skbuff won't be dropped, but returned to
sockets's queue. Technically instead of 'skb_dequeue()', 'skb_peek()' is
called and when skbuff becomes empty, it is removed from queue by
'__skb_unlink()'.
Fixes: 71dc9ec9ac ("virtio/vsock: replace virtio_vsock_pkt with sk_buff")
Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Acked-by: Bobby Eshleman <bobby.eshleman@bytedance.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since we now no longer use 'skb->len' to update credit, there is no sense
to update skbuff state, because it is used only once after dequeue to
copy data and then will be released.
Fixes: 71dc9ec9ac ("virtio/vsock: replace virtio_vsock_pkt with sk_buff")
Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Acked-by: Bobby Eshleman <bobby.eshleman@bytedance.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
'skb->len' can vary when we partially read the data, this complicates the
calculation of credit to be updated in 'virtio_transport_inc_rx_pkt()/
virtio_transport_dec_rx_pkt()'.
Also in 'virtio_transport_dec_rx_pkt()' we were miscalculating the
credit since 'skb->len' was redundant.
For these reasons, let's replace the use of skbuff state to calculate new
'rx_bytes'/'fwd_cnt' values with explicit value as input argument. This
makes code more simple, because it is not needed to change skbuff state
before each call to update 'rx_bytes'/'fwd_cnt'.
Fixes: 71dc9ec9ac ("virtio/vsock: replace virtio_vsock_pkt with sk_buff")
Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Acked-by: Bobby Eshleman <bobby.eshleman@bytedance.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This commit changes virtio/vsock to use sk_buff instead of
virtio_vsock_pkt. Beyond better conforming to other net code, using
sk_buff allows vsock to use sk_buff-dependent features in the future
(such as sockmap) and improves throughput.
This patch introduces the following performance changes:
Tool: Uperf
Env: Phys Host + L1 Guest
Payload: 64k
Threads: 16
Test Runs: 10
Type: SOCK_STREAM
Before: commit b7bfaa761d ("Linux 6.2-rc3")
Before
------
g2h: 16.77Gb/s
h2g: 10.56Gb/s
After
-----
g2h: 21.04Gb/s
h2g: 10.76Gb/s
Signed-off-by: Bobby Eshleman <bobby.eshleman@bytedance.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This adds extra condition to wake up data reader: do it only when number
of readable bytes >= SO_RCVLOWAT. Otherwise, there is no sense to kick
user,because it will wait until SO_RCVLOWAT bytes will be dequeued. This
check is performed in vsock_data_ready().
Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This callback controls setting of POLLIN, POLLRDNORM output bits of poll()
syscall, but in some cases, it is incorrectly to set it, when socket has
at least 1 bytes of available data. Use 'target' which is already exists.
Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The VMADDR_CID_ANY flag used by a socket means that the socket isn't bound
to any specific CID. For example, a host vsock server may want to be bound
with VMADDR_CID_ANY, so that a guest vsock client can connect to the host
server with CID=VMADDR_CID_HOST (i.e. 2), and meanwhile, a host vsock
client can connect to the same local server with CID=VMADDR_CID_LOCAL
(i.e. 1).
The current implementation sets the destination socket's svm_cid to a
fixed CID value after the first client's connection, which isn't an
expected operation. For example, if the guest client first connects to the
host server, the server's svm_cid gets set to VMADDR_CID_HOST, then other
host clients won't be able to connect to the server anymore.
Reproduce steps:
1. Run the host server:
socat VSOCK-LISTEN:1234,fork -
2. Run a guest client to connect to the host server:
socat - VSOCK-CONNECT:2:1234
3. Run a host client to connect to the host server:
socat - VSOCK-CONNECT:1:1234
Without this patch, step 3. above fails to connect, and socat complains
"socat[1720] E connect(5, AF=40 cid:1 port:1234, 16): Connection
reset by peer".
With this patch, the above works well.
Fixes: c0cfa2d8a7 ("vsock: add multi-transports support")
Signed-off-by: Wei Wang <wei.w.wang@intel.com>
Link: https://lore.kernel.org/r/20211126011823.1760-1-wei.w.wang@intel.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
This current implemented bit is used to mark end of messages
('EOM' - end of message), not records('EOR' - end of record).
Also rename 'record' to 'message' in implementation as it is
different things.
Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Link: https://lore.kernel.org/r/20210903123109.3273053-1-arseny.krasnov@kaspersky.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
The original implementation of the virtio-vsock driver does not
handle a VIRTIO_VSOCK_OP_CREDIT_REQUEST as required by the
virtio-vsock specification. The vsock device emulated by
vhost-vsock and the virtio-vsock driver never uses this request,
which was probably why nobody noticed it. However, another
implementation of the device may use this request type.
Hence, this commit introduces a way to handle an explicit credit
request by responding with a corresponding credit update as
required by the virtio-vsock specification.
Fixes: 06a8fc7836 ("VSOCK: Introduce virtio_vsock_common.ko")
Signed-off-by: Harshavardhan Unnibhavi <harshanavkis@gmail.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Link: https://lore.kernel.org/r/20210802173506.2383-1-harshanavkis@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This patch introduces a function wrapper to call the sk_error_report
callback. That will prepare to add additional handling whenever
sk_error_report is called, for example to trace socket errors.
Signed-off-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When memcpy_to_msg() fails in virtio_transport_seqpacket_do_dequeue(),
we already set `dequeued_len` with the negative error value returned
by memcpy_to_msg().
So we can directly check `dequeued_len` value instead of using a
dedicated flag variable to skip the copy path for the rest of
fragments.
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Small updates to make SOCK_SEQPACKET work:
1) Send SHUTDOWN on socket close for SEQPACKET type.
2) Set SEQPACKET packet type during send.
3) Set 'VIRTIO_VSOCK_SEQ_EOR' bit in flags for last
packet of message.
4) Implement data check function for SEQPACKET.
5) Check for max datagram size.
Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Update current receive logic for SEQPACKET support: performs
check for packet and socket types on receive(if mismatch, then
reset connection). Increment EOR counter on receive. Also if
buffer of new packet was appended to buffer of last packet in
rx queue, update flags of last packet with flags of new packet.
Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Callback fetches RW packets from rx queue of socket until whole record
is copied(if user's buffer is full, user is not woken up). This is done
to not stall sender, because if we wake up user and it leaves syscall,
nobody will send credit update for rest of record, and sender will wait
for next enter of read syscall at receiver's side. So if user buffer is
full, we just send credit update and drop data.
Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This function is static and 'hdr' arg was always NULL.
Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is no need to set type of packet which differs from type
of socket, so move passing type of packet from 'info' structure
to 'virtio_transport_send_pkt_info()' function. Since at current
time only stream type is supported, set it directly in 'virtio_
transport_send_pkt_info()', so callers don't need to set it.
Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As reported by syzbot [1], there is a memory leak while closing the
socket. We partially solved this issue with commit ac03046ece
("vsock/virtio: free packets during the socket release"), but we
forgot to drain the RX queue when the socket is definitely closed by
the scheduled work.
To avoid future issues, let's use the new virtio_transport_remove_sock()
to drain the RX queue before removing the socket from the af_vsock lists
calling vsock_remove_sock().
[1] https://syzkaller.appspot.com/bug?extid=24452624fc4c571eedd9
Fixes: ac03046ece ("vsock/virtio: free packets during the socket release")
Reported-and-tested-by: syzbot+24452624fc4c571eedd9@syzkaller.appspotmail.com
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If the socket is closed or is being released, some resources used by
virtio_transport_space_update() such as 'vsk->trans' may be released.
To avoid a use after free bug we should only update the available credit
when we are sure the socket is still open and we have the lock held.
Fixes: 06a8fc7836 ("VSOCK: Introduce virtio_vsock_common.ko")
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Link: https://lore.kernel.org/r/20210208144454.84438-1-sgarzare@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Starting from commit 8692cefc43 ("virtio_vsock: Fix race condition
in virtio_transport_recv_pkt"), we discard packets in
virtio_transport_recv_pkt() if the socket has been released.
When the socket is connected, we schedule a delayed work to wait the
RST packet from the other peer, also if SHUTDOWN_MASK is set in
sk->sk_shutdown.
This is done to complete the virtio-vsock shutdown algorithm, releasing
the port assigned to the socket definitively only when the other peer
has consumed all the packets.
If we discard the RST packet received, the socket will be closed only
when the VSOCK_CLOSE_TIMEOUT is reached.
Sergio discovered the issue while running ab(1) HTTP benchmark using
libkrun [1] and observing a latency increase with that commit.
To avoid this issue, we discard packet only if the socket is really
closed (SOCK_DONE flag is set).
We also set SOCK_DONE in virtio_transport_release() when we don't need
to wait any packets from the other peer (we didn't schedule the delayed
work). In this case we remove the socket from the vsock lists, releasing
the port assigned.
[1] https://github.com/containers/libkrun
Fixes: 8692cefc43 ("virtio_vsock: Fix race condition in virtio_transport_recv_pkt")
Cc: justin.he@arm.com
Reported-by: Sergio Lopez <slp@redhat.com>
Tested-by: Sergio Lopez <slp@redhat.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Acked-by: Jia He <justin.he@arm.com>
Link: https://lore.kernel.org/r/20201120104736.73749-1-sgarzare@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When client on the host tries to connect(SOCK_STREAM, O_NONBLOCK) to the
server on the guest, there will be a panic on a ThunderX2 (armv8a server):
[ 463.718844] Unable to handle kernel NULL pointer dereference at virtual address 0000000000000000
[ 463.718848] Mem abort info:
[ 463.718849] ESR = 0x96000044
[ 463.718852] EC = 0x25: DABT (current EL), IL = 32 bits
[ 463.718853] SET = 0, FnV = 0
[ 463.718854] EA = 0, S1PTW = 0
[ 463.718855] Data abort info:
[ 463.718856] ISV = 0, ISS = 0x00000044
[ 463.718857] CM = 0, WnR = 1
[ 463.718859] user pgtable: 4k pages, 48-bit VAs, pgdp=0000008f6f6e9000
[ 463.718861] [0000000000000000] pgd=0000000000000000
[ 463.718866] Internal error: Oops: 96000044 [#1] SMP
[...]
[ 463.718977] CPU: 213 PID: 5040 Comm: vhost-5032 Tainted: G O 5.7.0-rc7+ #139
[ 463.718980] Hardware name: GIGABYTE R281-T91-00/MT91-FS1-00, BIOS F06 09/25/2018
[ 463.718982] pstate: 60400009 (nZCv daif +PAN -UAO)
[ 463.718995] pc : virtio_transport_recv_pkt+0x4c8/0xd40 [vmw_vsock_virtio_transport_common]
[ 463.718999] lr : virtio_transport_recv_pkt+0x1fc/0xd40 [vmw_vsock_virtio_transport_common]
[ 463.719000] sp : ffff80002dbe3c40
[...]
[ 463.719025] Call trace:
[ 463.719030] virtio_transport_recv_pkt+0x4c8/0xd40 [vmw_vsock_virtio_transport_common]
[ 463.719034] vhost_vsock_handle_tx_kick+0x360/0x408 [vhost_vsock]
[ 463.719041] vhost_worker+0x100/0x1a0 [vhost]
[ 463.719048] kthread+0x128/0x130
[ 463.719052] ret_from_fork+0x10/0x18
The race condition is as follows:
Task1 Task2
===== =====
__sock_release virtio_transport_recv_pkt
__vsock_release vsock_find_bound_socket (found sk)
lock_sock_nested
vsock_remove_sock
sock_orphan
sk_set_socket(sk, NULL)
sk->sk_shutdown = SHUTDOWN_MASK
...
release_sock
lock_sock
virtio_transport_recv_connecting
sk->sk_socket->state (panic!)
The root cause is that vsock_find_bound_socket can't hold the lock_sock,
so there is a small race window between vsock_find_bound_socket() and
lock_sock(). If __vsock_release() is running in another task,
sk->sk_socket will be set to NULL inadvertently.
This fixes it by checking sk->sk_shutdown(suggested by Stefano) after
lock_sock since sk->sk_shutdown is set to SHUTDOWN_MASK under the
protection of lock_sock_nested.
Signed-off-by: Jia He <justin.he@arm.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In virtio_transport.c, if the virtqueue is full, the transmitting
packet is queued up and it will be sent in the next iteration.
This causes the same packet to be delivered multiple times to
monitoring devices.
We want to continue to deliver packets to monitoring devices before
it is put in the virtqueue, to avoid that replies can appear in the
packet capture before the transmitted packet.
This patch fixes the issue, adding a new flag (tap_delivered) in
struct virtio_vsock_pkt, to check if the packet is already delivered
to monitoring devices.
In vhost/vsock.c, we are splitting packets, so we must set
'tap_delivered' to false when we queue up the same virtio_vsock_pkt
to handle the remaining bytes.
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The mptcp conflict was overlapping additions.
The SMC conflict was an additional and removal happening at the same
time.
Signed-off-by: David S. Miller <davem@davemloft.net>
Some transports (hyperv, virtio) acquire the sock lock during the
.release() callback.
In the vsock_stream_connect() we call vsock_assign_transport(); if
the socket was previously assigned to another transport, the
vsk->transport->release() is called, but the sock lock is already
held in the vsock_stream_connect(), causing a deadlock reported by
syzbot:
INFO: task syz-executor280:9768 blocked for more than 143 seconds.
Not tainted 5.6.0-rc1-syzkaller #0
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
syz-executor280 D27912 9768 9766 0x00000000
Call Trace:
context_switch kernel/sched/core.c:3386 [inline]
__schedule+0x934/0x1f90 kernel/sched/core.c:4082
schedule+0xdc/0x2b0 kernel/sched/core.c:4156
__lock_sock+0x165/0x290 net/core/sock.c:2413
lock_sock_nested+0xfe/0x120 net/core/sock.c:2938
virtio_transport_release+0xc4/0xd60 net/vmw_vsock/virtio_transport_common.c:832
vsock_assign_transport+0xf3/0x3b0 net/vmw_vsock/af_vsock.c:454
vsock_stream_connect+0x2b3/0xc70 net/vmw_vsock/af_vsock.c:1288
__sys_connect_file+0x161/0x1c0 net/socket.c:1857
__sys_connect+0x174/0x1b0 net/socket.c:1874
__do_sys_connect net/socket.c:1885 [inline]
__se_sys_connect net/socket.c:1882 [inline]
__x64_sys_connect+0x73/0xb0 net/socket.c:1882
do_syscall_64+0xfa/0x790 arch/x86/entry/common.c:294
entry_SYSCALL_64_after_hwframe+0x49/0xbe
To avoid this issue, this patch remove the lock acquiring in the
.release() callback of hyperv and virtio transports, and it holds
the lock when we call vsk->transport->release() in the vsock core.
Reported-by: syzbot+731710996d79d0d58fbc@syzkaller.appspotmail.com
Fixes: 408624af4c ("vsock: use local transport when it is loaded")
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Whenever the vsock backend on the host sends a packet through the RX
queue, it expects an answer on the TX queue. Unfortunately, there is one
case where the host side will hang waiting for the answer and might
effectively never recover if no timeout mechanism was implemented.
This issue happens when the guest side starts binding to the socket,
which insert a new bound socket into the list of already bound sockets.
At this time, we expect the guest to also start listening, which will
trigger the sk_state to move from TCP_CLOSE to TCP_LISTEN. The problem
occurs if the host side queued a RX packet and triggered an interrupt
right between the end of the binding process and the beginning of the
listening process. In this specific case, the function processing the
packet virtio_transport_recv_pkt() will find a bound socket, which means
it will hit the switch statement checking for the sk_state, but the
state won't be changed into TCP_LISTEN yet, which leads the code to pick
the default statement. This default statement will only free the buffer,
while it should also respond to the host side, by sending a packet on
its TX queue.
In order to simply fix this unfortunate chain of events, it is important
that in case the default statement is entered, and because at this stage
we know the host side is waiting for an answer, we must send back a
packet containing the operation VIRTIO_VSOCK_OP_RST.
One could say that a proper timeout mechanism on the host side will be
enough to avoid the backend to hang. But the point of this patch is to
ensure the normal use case will be provided with proper responsiveness
when it comes to establishing the connection.
Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
virtio_transport_get_ops() and virtio_transport_send_pkt_info()
can only be used on connecting/connected sockets, since a socket
assigned to a transport is required.
This patch adds a WARN_ON() on virtio_transport_get_ops() to check
this requirement, a comment and a returned error on
virtio_transport_send_pkt_info(),
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We can remove virtio header includes, because virtio_transport_common
doesn't use virtio API, but provides common functions to interface
virtio/vhost transports with the af_vsock core, and to handle
the protocol.
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds the support of multiple transports in the
VSOCK core.
With the multi-transports support, we can use vsock with nested VMs
(using also different hypervisors) loading both guest->host and
host->guest transports at the same time.
Major changes:
- vsock core module can be loaded regardless of the transports
- vsock_core_init() and vsock_core_exit() are renamed to
vsock_core_register() and vsock_core_unregister()
- vsock_core_register() has a feature parameter (H2G, G2H, DGRAM)
to identify which directions the transport can handle and if it's
support DGRAM (only vmci)
- each stream socket is assigned to a transport when the remote CID
is set (during the connect() or when we receive a connection request
on a listener socket).
The remote CID is used to decide which transport to use:
- remote CID <= VMADDR_CID_HOST will use guest->host transport;
- remote CID == local_cid (guest->host transport) will use guest->host
transport for loopback (host->guest transports don't support loopback);
- remote CID > VMADDR_CID_HOST will use host->guest transport;
- listener sockets are not bound to any transports since no transport
operations are done on it. In this way we can create a listener
socket, also if the transports are not loaded or with VMADDR_CID_ANY
to listen on all transports.
- DGRAM sockets are handled as before, since only the vmci_transport
provides this feature.
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
All transports call __vsock_create() with the same parameters,
most of them depending on the parent socket. In order to simplify
the VSOCK core APIs exposed to the transports, this patch adds
the vsock_create_connected() callable from transports to create
a new socket when a connection request is received.
We also unexported the __vsock_create().
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
virtio_transport and vmci_transport handle the buffer_size
sockopts in a very similar way.
In order to support multiple transports, this patch moves this
handling in the core to allow the user to change the options
also if the socket is not yet assigned to any transport.
This patch also adds the '.notify_buffer_size' callback in the
'struct virtio_transport' in order to inform the transport,
when the buffer_size is changed by the user. It is also useful
to limit the 'buffer_size' requested (e.g. virtio transports).
Acked-by: Dexuan Cui <decui@microsoft.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since now the 'struct vsock_sock' object contains a pointer to
the transport, this patch adds a parameter to the
vsock_core_get_transport() to return the right transport
assigned to the socket.
This patch modifies also the virtio_transport_get_ops(), that
uses the vsock_core_get_transport(), adding the
'struct vsock_sock *' parameter.
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
